Preprint reports iterative framework lifts coding-agent benchmark results

A preprint describes Harness-of-Harness, a framework that organizes coding-agent harnesses into iterative planning, coding and testing loops intended to improve autonomous software development. The authors report average relative gains of 52.25 percent, with a maximum gain of 82.86 percent after three iterations, across GameCraft-Bench, FrontierSWE and ProgramBench for three harness-model pairs. They also describe a deployment of more than 70 iterations that produced a playable first-person-shooter game with implemented mechanics, visuals and audio.